Using Large Language Models to Software Requirements Selection for Scalable, Explainable, and Reliable Results

The software requirements selection (SRS) is one of the primary activities that decides the success or failure scenarios of software projects. Conventional approaches adopted for the SRS process had several issues, such as bias, limitations of scaling, and absence of clarity. To tackle these limitations, this paper provides a strong integration of the large language models (LLMs) into the SRS process. With the help of the LLMs, it is possible to automate and enhance the process of performing tasks like requirement analysis, requirements prioritization, and decision-making. The proposed LLM-based framework leverages the semantic understanding of LLMs. It analyzes the stakeholders' inputs, learns from historical data, and considers existing project constraints to support more precise and efficient requirements handling. The security and explainability concerns of using LLMs in decision-making scenarios are also examined in this paper. Furthermore, the issue of reliability is also addressed to ensure consistency, robustness, and reproducibility of the LLM-driven decisions.